From mboxrd@z Thu Jan 1 00:00:00 1970 From: Paul Clements Subject: Re: Bug report: mdadm -E oddity Date: Fri, 20 May 2005 16:05:11 -0400 Message-ID: <428E42F7.40208@steeleye.com> References: <1115999051.3974.14.camel@compaq-rhel4.xsintricity.com> <1116004267.3974.35.camel@compaq-rhel4.xsintricity.com> <17029.12773.197506.463977@cse.unsw.edu.au> <1116077316.13780.52.camel@compaq-rhel4.xsintricity.com> <17037.35615.456231.737766@cse.unsw.edu.au> <1116592212.23785.75.camel@compaq-rhel4.xsintricity.com> <428E0A92.3060108@steeleye.com> <1116611115.23785.89.camel@compaq-rhel4.xsintricity.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: Sender: linux-raid-owner@vger.kernel.org To: linux-raid@vger.kernel.org List-Id: linux-raid.ids Peter T. Breuer wrote: > b) rewriting is not necessarily idempotent, when half of it consists > of using a parity to construct what you should write. That's right. The rewrite is not idempotent. Having lost part, or all, of either D1 or P at the time of the crash, you no longer have any accurate way of reconstructing D2. And that's assuming you can even do a rewrite. By default, only metadata is journalled (in ext3), so for plain old data writes, you're just plain out of luck... -- Paul